WEBVTT
00:00:03.600 --> 00:00:11.560
The Riemann zeta function, this is one of those objects in modern math that a lot of you might have heard of, but which can be really difficult to understand.
00:00:12.280 --> 00:00:15.200
Don’t worry, I’ll explain that animation that you just saw in a few minutes.
00:00:15.960 --> 00:00:22.120
A lot of people know about this function because there’s a one-million-dollar prize out for anyone who can figure out when it equals zero.
00:00:22.720 --> 00:00:25.200
An open problem known as the Riemann hypothesis.
00:00:26.000 --> 00:00:32.600
Some of you may have heard of it in the context of the divergent sum one plus two plus three plus four, on and on up to infinity.
00:00:33.240 --> 00:00:40.080
You see, there’s a sense in which the sum equals negative one twelfth, which seems nonsensical if not obviously wrong.
00:00:40.880 --> 00:00:45.920
But a common way to define what this equation is actually saying uses the Riemann zeta function.
00:00:46.880 --> 00:00:57.440
But as any casual math enthusiast who started to read into this knows, its definition references this one idea called analytic continuation, which has to do with complex-valued functions.
00:00:57.840 --> 00:01:00.440
And this idea can be frustratingly opaque and unintuitive.
00:01:01.400 --> 00:01:11.640
So what I’d like to do here is just show you all what this zeta function actually looks like and to explain what this idea of analytic continuation is in a visual and more intuitive way.
00:01:14.240 --> 00:01:17.880
I’m assuming that you know about complex numbers and that you’re comfortable working with them.
00:01:18.520 --> 00:01:24.040
And I’m tempted to say that you should know calculus since analytic continuation is all about derivatives.
00:01:24.320 --> 00:01:28.080
But for the way I’m planning to present things, I think you might actually be fine without that.
00:01:29.160 --> 00:01:32.840
So to jump right into it, let’s just define what this zeta function is.
00:01:33.400 --> 00:01:48.800
For a given input, where we commonly use the variable 𝑠, the function is one over one to the 𝑠, which is always one, plus one over two to the 𝑠 plus one over three to the 𝑠 plus one over four to the 𝑠, on and on and on, summing up over all natural numbers.
00:01:51.120 --> 00:01:55.280
So, for example, let’s say you plug in a value like 𝑠 equals two.
00:01:55.600 --> 00:02:00.280
You’d get one plus one over four plus one over nine plus one sixteenth.
00:02:00.840 --> 00:02:09.280
And as you keep adding more and more reciprocals of squares, this just so happens to approach 𝜋 squared over six, which is around 1.645.
00:02:09.960 --> 00:02:12.560
There’s a very beautiful reason for why 𝜋 shows up here.
00:02:12.560 --> 00:02:14.280
And I might do a video on it at a later day.
00:02:14.640 --> 00:02:17.800
But that’s just the tip of the iceberg for why this function is beautiful.
00:02:18.800 --> 00:02:22.200
You could do the same thing for other inputs 𝑠, like three or four.
00:02:22.520 --> 00:02:24.480
And sometimes you get other interesting values.
00:02:25.160 --> 00:02:27.240
And so far, everything feels pretty reasonable.
00:02:27.680 --> 00:02:31.560
You’re adding up smaller and smaller amounts, and these sums approach some number.
00:02:32.160 --> 00:02:33.400
Great, no craziness here.
00:02:34.520 --> 00:02:40.440
Yet, if you were to read about it, you might see some people say that zeta of negative one equals negative one twelfth.
00:02:41.600 --> 00:02:44.480
But looking at this infinite sum, that doesn’t make any sense.
00:02:45.120 --> 00:02:55.240
When you raise each term to the negative one, flipping each fraction, you get one plus two plus three plus four, on and on, over all natural numbers.
00:02:55.640 --> 00:02:59.720
And obviously that doesn’t approach anything, certainly not negative one twelfth, right?
00:03:01.120 --> 00:03:08.880
And, as any mercenary looking into the Riemann hypothesis knows, this function is said to have trivial zeros at negative even numbers.
00:03:09.360 --> 00:03:12.720
So, for example, that would mean that zeta of negative two equals zero.
00:03:13.520 --> 00:03:23.720
But when you plug in negative two, it gives you one plus four plus nine plus 16, on and on, which again obviously doesn’t approach anything, much less zero, right?
00:03:25.040 --> 00:03:27.320
Well, we’ll get to negative values in a few minutes.
00:03:27.320 --> 00:03:30.640
But for right now, let’s just say the only thing that seems reasonable.
00:03:31.200 --> 00:03:36.280
This function only makes sense when 𝑠 is greater than one, which is when this sum converges.
00:03:36.800 --> 00:03:39.840
So far, it’s simply not defined for other values.
00:03:41.200 --> 00:03:49.840
Now with that said, Bernhard Riemann was somewhat of a father to complex analysis, which is the study of functions that have complex numbers as inputs and outputs.
00:03:50.680 --> 00:04:03.280
So rather than just thinking about how this sum takes a number 𝑠 on the real number line to another number on the real number line, his main focus was on understanding what happens when you plug in a complex value for 𝑠.
00:04:04.120 --> 00:04:08.440
So, for example, maybe instead of plugging in two, you would plug in two plus 𝑖.
00:04:10.560 --> 00:04:16.320
Now if you’ve never seen the idea of raising a number to the power of a complex value, you can feel kind of strange at first.
00:04:16.520 --> 00:04:19.720
Because it no longer has anything to do with repeated multiplication.
00:04:20.600 --> 00:04:31.000
But mathematicians found that there is a very nice and very natural way to extend the definition of exponents beyond their familiar territory of real numbers and into the realm of complex values.
00:04:33.200 --> 00:04:37.360
It’s not super crucial to understand complex exponents for where I’m going with this video.
00:04:37.880 --> 00:04:40.800
But I think it’ll still be nice if we just summarize the gist of it here.
00:04:41.440 --> 00:04:52.440
The basic idea is that when you write something like one-half to the power of a complex number, you split it up as one-half to the real part times one-half to the pure imaginary part.
00:04:52.440 --> 00:04:55.280
We’re good on one-half to the real part; there’s no issues there.
00:04:55.280 --> 00:04:58.680
But what about raising something to a pure imaginary number?
00:05:01.560 --> 00:05:08.360
Well, the result is gonna be some complex number on the unit circle in the complex plane.
00:05:09.480 --> 00:05:17.280
As you let that pure imaginary input walk up and down the imaginary line, the resulting output walks around that unit circle.
00:05:21.680 --> 00:05:25.760
For a base like one-half, the output walks around the unit circle somewhat slowly.
00:05:27.120 --> 00:05:38.160
But for a base that’s farther away from one, like one-ninth, then as you let this input walk up and down the imaginary axis, the corresponding output is gonna walk around the unit circle more quickly.
00:05:39.280 --> 00:05:44.800
If you’ve never seen this and you’re wondering why on earth this happens, I’ve left a few links to good resources in the description.
00:05:45.320 --> 00:05:48.000
For here, I’m just gonna move forward with the what without the why.
00:05:48.000 --> 00:05:57.280
The main takeaway is that when you raise something like one-half to the power of two plus 𝑖, which is one-half squared times one-half to the 𝑖.
00:05:57.840 --> 00:06:03.720
That one-half to the 𝑖 part is gonna be on the unit circle, meaning it has an absolute value of one.
00:06:05.920 --> 00:06:12.000
So when you multiply it, it doesn’t change the size of the number; it just takes that one-fourth and rotates it somewhat.
00:06:15.320 --> 00:06:24.480
So if you were to plug in two plus 𝑖 to the zeta function, one way to think about what it does is to start off with all of the terms raised to the power of two.
00:06:24.480 --> 00:06:33.600
Which you can think of as piecing together lines whose lengths are the reciprocals of squares of numbers which, like I said before, converges to 𝜋 squared over six.
00:06:34.280 --> 00:06:40.240
Then when you change that input from two up to two plus 𝑖, each of these lines gets rotated by some amount.
00:06:41.080 --> 00:06:46.080
But importantly, the lengths of those lines won’t change, so the sum still converges.
00:06:46.080 --> 00:06:49.480
It just does so in a spiral to some specific point on the complex plane.
00:06:50.840 --> 00:06:56.480
Here, let me show what it looks like when I vary the input 𝑠 represented with this yellow dot on the complex plane.
00:06:56.800 --> 00:07:06.960
Where this spiral sum is always gonna be showing the converging value for zeta of 𝑠.
00:07:13.240 --> 00:07:23.240
What this means is that zeta of 𝑠, defined as this infinite sum, is a perfectly reasonable complex function as long as the real part of the input is greater than one.
00:07:23.800 --> 00:07:28.280
Meaning, the input 𝑠 sits somewhere on this right half of the complex plane.
00:07:29.040 --> 00:07:38.240
Again, this is because it’s the real part of 𝑠 that determines the size of each number, while the imaginary part just dictates some rotation.
00:07:39.520 --> 00:07:42.320
So now what I wanna do is visualize this function.
00:07:42.880 --> 00:07:48.920
It takes in inputs on the right half of the complex plane and spits out outputs somewhere else in the complex plane.
00:07:49.840 --> 00:07:55.120
A super nice way to understand complex functions is to visualize them as transformations.
00:07:55.120 --> 00:08:01.080
Meaning, you look at every possible input to the function and just let it move over to the corresponding output.
00:08:02.040 --> 00:08:08.840
For example, let’s take a moment and try to visualize something a little bit easier than the zeta function, say 𝑓 of 𝑠 is equal to 𝑠 squared.
00:08:09.720 --> 00:08:11.960
When you plug in 𝑠 equals two, you get four.
00:08:12.440 --> 00:08:16.120
So we’ll end up moving that point at two over to the point at four.
00:08:16.960 --> 00:08:19.320
When you plug in negative one, you get one.
00:08:19.680 --> 00:08:23.920
So the point over here at negative one is gonna end up moving over to the point at one.
00:08:25.160 --> 00:08:31.200
When you plug in 𝑖, by definition, its square is negative one, so it’s gonna move over here to negative one.
00:08:32.480 --> 00:08:34.640
Now I’m gonna add on a more colorful grid.
00:08:34.640 --> 00:08:36.800
And this is just because things are about to start moving.
00:08:36.800 --> 00:08:40.240
And it’s kinda nice to have something to distinguish grid lines during that movement.
00:08:41.000 --> 00:08:50.840
From here, I’ll tell the computer to move every single point on this grid over to its corresponding output under the function 𝑓 of 𝑠 equals 𝑠 squared.
00:08:50.840 --> 00:08:51.000
Here’s what it looks like.
00:08:55.760 --> 00:08:58.160
That can be a lot to take in, so I’ll go ahead and play it again.
00:08:58.800 --> 00:09:01.240
And this time, focus on one of the marked points.
00:09:01.440 --> 00:09:04.920
And notice how it moves over to the point corresponding to its square.
00:09:07.240 --> 00:09:10.720
It can be a little complicated to see all of the points moving all at once.
00:09:10.960 --> 00:09:16.120
But the reward is that this gives us a very rich picture for what the complex function is actually doing.
00:09:16.600 --> 00:09:18.280
And it all happens in just two dimensions.
00:09:20.520 --> 00:09:25.880
So back to the zeta function, we have this infinite sum, which is a function of some complex number 𝑠.
00:09:26.200 --> 00:09:30.720
And we feel good and happy about plugging in values of 𝑠 whose real part is greater than one.
00:09:31.440 --> 00:09:34.640
And getting some meaningful output via the converging spiral sum.
00:09:35.680 --> 00:09:43.920
So to visualize this function, I’m gonna take the portion of the grid sitting on the right side of the complex plane here, where the real part of numbers is greater than one.
00:09:44.600 --> 00:09:48.400
And I’m gonna tell the computer to move each point of this grid to the appropriate output.
00:09:49.280 --> 00:09:52.480
It actually helps if I add a few more grid lines around the number one.
00:09:52.480 --> 00:09:54.600
Since that region gets stretched out by quite a bit.
00:09:59.920 --> 00:10:03.560
Alright, so first of all, let’s all just appreciate how beautiful that is.
00:10:03.920 --> 00:10:08.960
I mean, damn, that doesn’t make you wanna learn more about complex functions, you have no heart.
00:10:10.840 --> 00:10:15.520
But also, this transformed grid is just begging to be extended a little bit.
00:10:16.200 --> 00:10:24.600
For example, let’s highlight these lines here, which represent all of the complex numbers with imaginary part 𝑖 or negative 𝑖.
00:10:27.240 --> 00:10:32.160
After the transformation, these lines make such lovely arcs before they just abruptly stop.
00:10:32.920 --> 00:10:35.880
Don’t you wanna just, you know, continue those arcs?
00:10:36.760 --> 00:10:47.160
In fact, you can imagine how some altered version of the function with a definition that extends into this left half of the plane might be able to complete this picture with something that’s quite pretty.
00:10:48.240 --> 00:10:52.280
Well, this is exactly what mathematicians working with complex functions do.
00:10:52.960 --> 00:10:57.240
They continue the function beyond the original domain where it was defined.
00:10:58.320 --> 00:11:07.120
Now as soon as we branch over into inputs where the real part is less than one, this infinite sum that we originally used to define the function doesn’t make sense anymore.
00:11:07.520 --> 00:11:11.520
You’ll get nonsense like adding one plus two plus three plus four, on and on up to infinity.
00:11:12.320 --> 00:11:17.360
But just looking at this transformed version of the right half of the plane where the sum does make sense.
00:11:17.840 --> 00:11:21.880
It’s just begging us to extend the set of points that we’re considering as inputs.
00:11:22.400 --> 00:11:27.920
Even if that means defining the extended function in some way that doesn’t necessarily use that sum.
00:11:29.120 --> 00:11:33.160
Of course, that leaves us with the question, how would you define that function on the rest of the plane?
00:11:34.960 --> 00:11:37.720
You might think that you could extend it any number of ways.
00:11:38.240 --> 00:11:44.680
Maybe you define an extension that makes it so the point at, say, 𝑠 equals negative one moves over to negative one twelfth.
00:11:47.880 --> 00:11:51.240
But maybe you squiggle on some extension that makes it land on any other value.
00:11:52.000 --> 00:12:03.520
I mean, as soon as you open yourself up to the idea of defining the function differently for values outside that domain of convergence — that is, not based on this infinite sum — the world is your oyster.
00:12:03.760 --> 00:12:06.240
And you can have any number of extensions, right?
00:12:07.560 --> 00:12:08.920
Well, not exactly.
00:12:09.400 --> 00:12:13.920
I mean yes, you can give any child a marker and have them extend these lines any which way.
00:12:14.520 --> 00:12:23.720
But if you add on the restriction that this new extended function has to have a derivative everywhere, it locks us into one and only one possible extension.
00:12:25.280 --> 00:12:28.720
I know, I know, I said that you wouldn’t need to know about derivatives for this video.
00:12:29.160 --> 00:12:34.160
And even if you do know calculus, maybe you have yet to learn how to interpret derivatives for complex functions.
00:12:34.840 --> 00:12:42.320
But luckily for us, there is a very nice geometric intuition that you can keep in mind for when I say a phrase like “has a derivative everywhere.”
00:12:43.240 --> 00:12:47.280
Here, to show you what I mean, let’s look back at that 𝑓 of 𝑠 equals 𝑠 squared example.
00:12:48.040 --> 00:12:54.960
Again, we think of this function as a transformation moving every point 𝑠 of the complex plane over to the point 𝑠 squared.
00:12:56.040 --> 00:13:00.840
For those of you who know calculus, you know that you can take the derivative of this function at any given input.
00:13:01.600 --> 00:13:07.480
But there’s an interesting property of that transformation that turns out to be related and almost equivalent to that fact.
00:13:08.800 --> 00:13:19.400
If you look at any two lines in the input space that intersect at some angle and consider what they turn into after the transformation, they will still intersect each other at that same angle.
00:13:21.080 --> 00:13:22.880
The lines might get curved and that’s okay.
00:13:23.280 --> 00:13:27.760
But the important part is that the angle at which they intersect remains unchanged.
00:13:28.160 --> 00:13:31.040
And this is true for any pair of lines that you choose.
00:13:31.040 --> 00:13:40.400
So when I say a function has a derivative everywhere, I want you to think about this angle-preserving property.
00:13:41.000 --> 00:13:46.760
That anytime two lines intersect, the angle between them remains unchanged after the transformation.
00:13:47.840 --> 00:13:56.000
At a glance, this is easiest to appreciate by noticing how all of the curves that the gridlines turn into still intersect each other at right angles.
00:13:58.600 --> 00:14:02.200
Complex functions that have a derivative everywhere are called analytic.
00:14:02.640 --> 00:14:06.640
So you can think of this term analytic as meaning angle preserving.
00:14:06.640 --> 00:14:09.800
Admittedly, I’m lying to you a little here, but only a little bit.
00:14:10.360 --> 00:14:19.760
A slight caveat for those of you who want the full details is that at inputs where the derivative of a function is zero, instead of angles being preserved, they get multiplied by some integer.
00:14:20.680 --> 00:14:22.840
But those points are by far the minority.
00:14:23.160 --> 00:14:26.960
And for almost all inputs to an analytic function, angles are preserved.
00:14:26.960 --> 00:14:34.440
So if when I say analytic, you think angle preserving, I think that’s a fine intuition to have.
00:14:36.320 --> 00:14:45.720
Now if you think about it for a moment, and this is a point that I really want you to appreciate, this is a very restrictive property.
00:14:46.360 --> 00:14:50.720
The angle between any pair of intersecting lines has to remain unchanged.
00:14:51.600 --> 00:14:55.880
And yet, pretty much any function out there that has a name turns out to be analytic.
00:14:58.400 --> 00:15:10.720
The field of complex analysis, which Riemann helped to establish in its modern form, is almost entirely about leveraging the properties of analytic functions to understand the results and patterns in other fields of math and science.
00:15:12.920 --> 00:15:18.360
The zeta function defined by this infinite sum on the right half of the plane is an analytic function.
00:15:19.320 --> 00:15:24.680
Notice how all of these curves that the gridlines turn into still intersect each other at right angles.
00:15:28.000 --> 00:15:36.440
So the surprising fact about complex functions is that if you wanna extend an analytic function beyond the domain where it was originally defined.
00:15:37.080 --> 00:15:41.080
For example, extending this zeta function into the left half of the plane.
00:15:41.480 --> 00:15:52.880
Then if you require that the new extended function still be analytic — that is, that it still preserves angles everywhere — it forces you into only one possible extension, if one exists at all.
00:15:53.840 --> 00:16:03.880
It’s kind of like an infinite continuous jigsaw puzzle where this requirement of preserving angles walks you into one and only one choice for how to extend it.
00:16:04.360 --> 00:16:12.560
This process of extending an analytic function in the only way possible that’s still analytic is called, as you may have guessed, analytic continuation.
00:16:15.000 --> 00:16:17.600
So that’s how the full Riemann zeta function is defined.
00:16:18.320 --> 00:16:25.240
For values of 𝑠 on the right half of the plane, where the real part is greater than one, just plug them into this sum and see where it converges.
00:16:25.680 --> 00:16:27.920
And that convergence might look like some kind of spiral.
00:16:27.920 --> 00:16:33.040
Since raising each of these terms to a complex power has the effect of rotating each one.
00:16:33.560 --> 00:16:41.400
Then for the rest of the plane, we know that there exists one and only one way to extend this definition so that the function will still be analytic.
00:16:41.680 --> 00:16:44.760
That is, so that it still preserves angles at every single point.
00:16:45.280 --> 00:16:52.040
So we just say that, by definition, the zeta function on the left half of the plane is whatever that extension happens to be.
00:16:52.800 --> 00:16:57.520
And that’s a valid definition because there’s only one possible analytic continuation.
00:16:58.600 --> 00:17:00.920
Notice, that’s a very implicit definition.
00:17:01.360 --> 00:17:07.360
It just says, use the solution of this jigsaw puzzle, which, through more abstract derivation, we know must exist.
00:17:07.840 --> 00:17:10.240
But it doesn’t specify exactly how to solve it.
00:17:11.080 --> 00:17:14.320
Mathematicians have a pretty good grasp on what this extension looks like.
00:17:14.840 --> 00:17:19.120
But some important parts of it remain a mystery, a million-dollar mystery in fact.
00:17:19.680 --> 00:17:24.640
Let’s actually take a moment and talk about the Riemann hypothesis, the million-dollar problem.
00:17:24.960 --> 00:17:28.200
The places where this function equals zero turn out to be quite important.
00:17:28.880 --> 00:17:33.320
That is, which points get mapped onto the origin after the transformation?
00:17:34.400 --> 00:17:39.200
One thing we know about this extension is that the negative even numbers get mapped to zero.
00:17:41.560 --> 00:17:43.640
These are commonly called the trivial zeros.
00:17:44.280 --> 00:17:50.480
The naming here stems from a long-standing tradition of mathematicians to call things trivial when they understand it quite well.
00:17:50.680 --> 00:17:53.480
Even when it’s a fact that is not at all obvious from the outside.
00:17:54.840 --> 00:18:01.360
We also know that the rest of the points that get mapped to zero sit somewhere in this vertical strip, called the critical strip.
00:18:02.080 --> 00:18:08.200
And the specific placement of those nontrivial zeros encodes a surprising information about prime numbers.
00:18:09.120 --> 00:18:12.880
It’s actually pretty interesting why this function carries so much information about primes.
00:18:13.280 --> 00:18:15.600
And I definitely think I’ll make a video about that later on.
00:18:15.600 --> 00:18:18.800
But right now, things are long enough, so I’ll leave it unexplained.
00:18:19.680 --> 00:18:28.800
Riemann hypothesized that all of these nontrivial zeros sit right in the middle of the strip on the line of numbers 𝑠 whose real part is one-half.
00:18:29.480 --> 00:18:30.840
This is called the critical line.
00:18:31.720 --> 00:18:39.480
If that’s true, it gives us a remarkably tight grasp on the pattern of prime numbers as well as many other patterns in math that stem from this.
00:18:40.280 --> 00:18:47.000
Now so far, when I’ve shown what the zeta function looks like, I’ve only shown what it does to the portion of the grid on the screen.
00:18:47.600 --> 00:18:49.640
And that kind of undersells its complexity.
00:18:50.200 --> 00:18:56.440
So if I were to highlight this critical line and apply the transformation, it might not seem to cross the origin at all.
00:18:57.160 --> 00:19:01.280
However, here’s what the transformed version of more and more of that line looks like.
00:19:04.960 --> 00:19:09.880
Notice how it’s passing through the number zero many, many times.
00:19:10.840 --> 00:19:17.840
If you can prove that all of the nontrivial zeros sit somewhere on this line, the Clay Math Institute gives you one million dollars.
00:19:18.280 --> 00:19:25.320
And you’d also be proving hundreds if not thousands of modern math results that have already been shown contingent on this hypothesis being true.
00:19:26.480 --> 00:19:32.080
Another thing we know about this extended function is that it maps the point negative one over to negative one twelfth.
00:19:32.800 --> 00:19:42.200
And if you plug this into the original sum, it looks like we’re saying one plus two plus three plus four, on and on up to infinity, equals negative one twelfth.
00:19:42.720 --> 00:19:45.320
Now it might seem disingenuous to still call this a sum.
00:19:45.600 --> 00:19:51.080
Since the definition of the zeta function on the left half of the plane is not defined directly from this sum.
00:19:51.640 --> 00:19:56.640
Instead, it comes from analytically continuing the sum beyond the domain where it converges.
00:19:56.840 --> 00:20:01.040
That is, solving the jigsaw puzzle that began on the right half of the plane.
00:20:01.880 --> 00:20:14.880
That said, you have to admit that the uniqueness of this analytic continuation, the fact that the jigsaw puzzle has only one solution, is very suggestive of some intrinsic connection between these extended values and the original sum.
00:20:16.320 --> 00:20:35.280
For the last animation, and this is actually pretty cool, I’m gonna show you guys what the derivative of the zeta function looks like.